Dave framed this better than most: either the market sees through depreciation because the cash already left the building, or it doesn't. He built a model rather than picking a side.
Take the first framework seriously and the debate stops being about accounting. It becomes a question about asset life, which is observable.
CoreWeave disclosed this month a newly signed A100 contract running into 2029. The A100 launched in 2020 — nine years of commercial life. Jassy, the same week: servers break even in "a little less than 3 years," useful life "at least 5 to 6."
Oracle's 10-K says six years. Not five. Worth conceding — and then testing against what the chips actually do.
I discussed this problem in my latest Oracle ($ORCL) Deep Dive: The Fourth Hyperscaler, on Its Way to $1 Trillion.
There is an old banking adage about how to successfully underwrite risk. "People, Credit, Collateral". In the brave new world of Compute finance, underwriters are struggling with the "People" and "Collateral" parts of this axiom. As the reliable players are sorted from the overleveraged, spreads and terms will reflect the new knowledge. I suspect NVDA being willing to provide some level of residual value backstop to the new financing agreement is analogous to an automaker guaranteeing a minimum value of the cars supporting a pool of lease agreements they have originated through the dealership network and now are trying to sell. Over time the pool value will be established and the necessity for the backstop will diminish. Or NVDA will write a check.
You’re probably correct here, though I will note that jensen’s stated claims about gpu residual values (that they’re strong) is incongruent with nvidia serving as a backstop to those residual values.
Concur. In time we will probably learn how much risk they really believe they are assuming, vs reassuring skittish lenders. Coaxing a bunch of investment bankers into a half billion-dollar flyer took some creativity. By creating additional liquidity on NVDA balance sheet for higher value investments and establishing method to finance NVDA chip purchases (so far only NVDA chips) they are further separating from the pack.
Dave framed this better than most: either the market sees through depreciation because the cash already left the building, or it doesn't. He built a model rather than picking a side.
Take the first framework seriously and the debate stops being about accounting. It becomes a question about asset life, which is observable.
CoreWeave disclosed this month a newly signed A100 contract running into 2029. The A100 launched in 2020 — nine years of commercial life. Jassy, the same week: servers break even in "a little less than 3 years," useful life "at least 5 to 6."
Oracle's 10-K says six years. Not five. Worth conceding — and then testing against what the chips actually do.
I discussed this problem in my latest Oracle ($ORCL) Deep Dive: The Fourth Hyperscaler, on Its Way to $1 Trillion.
Thanks for the comment. To be clear, I had AI build a model, based on my instructions.
There is an old banking adage about how to successfully underwrite risk. "People, Credit, Collateral". In the brave new world of Compute finance, underwriters are struggling with the "People" and "Collateral" parts of this axiom. As the reliable players are sorted from the overleveraged, spreads and terms will reflect the new knowledge. I suspect NVDA being willing to provide some level of residual value backstop to the new financing agreement is analogous to an automaker guaranteeing a minimum value of the cars supporting a pool of lease agreements they have originated through the dealership network and now are trying to sell. Over time the pool value will be established and the necessity for the backstop will diminish. Or NVDA will write a check.
You’re probably correct here, though I will note that jensen’s stated claims about gpu residual values (that they’re strong) is incongruent with nvidia serving as a backstop to those residual values.
Concur. In time we will probably learn how much risk they really believe they are assuming, vs reassuring skittish lenders. Coaxing a bunch of investment bankers into a half billion-dollar flyer took some creativity. By creating additional liquidity on NVDA balance sheet for higher value investments and establishing method to finance NVDA chip purchases (so far only NVDA chips) they are further separating from the pack.
Great analysis as usual.